add support for compression body in http injection api

[dschaaff]

Compressing the body on HTTP injections can significantly save on bandwidth. It helps lower cross AZ bandwidth charged in AWS for example. I'll note that is my first time contribution in rust. I'm open to any changes or feedback there is. Thanks!

[wez]

Thanks for this!

I have an older unfinished branch from when I was initially experimenting with this, it looks similar but the layer is added more centrally:

diff --git a/crates/kumo-server-common/Cargo.toml b/crates/kumo-server-common/Cargo.toml
index 780dc087..16284260 100644
--- a/crates/kumo-server-common/Cargo.toml
+++ b/crates/kumo-server-common/Cargo.toml
@@ -60,7 +60,7 @@ throttle = {path="../throttle"}
 tokio = {workspace=true, features=["full", "tracing"]}
 tokio-metrics = "0.3.1"
 tokio-metrics-collector = "0.2.1"
-tower-http = {version="0.5", features=["trace", "compression-deflate", "compression-gzip"]}
+tower-http = {version="0.5", features=["trace", "compression-deflate", "compression-gzip", "decompression-deflate", "decompression-gzip"]}
 tracing = "0.1"
 tracing-appender = "0.2"
 tracing-subscriber = {version="0.3", features=["env-filter", "std", "fmt", "json"]}
diff --git a/crates/kumo-server-common/src/http_server/mod.rs b/crates/kumo-server-common/src/http_server/mod.rs
index 1bf5d293..4cffed4f 100644
--- a/crates/kumo-server-common/src/http_server/mod.rs
+++ b/crates/kumo-server-common/src/http_server/mod.rs
@@ -14,6 +14,7 @@ use serde::Deserialize;
 use std::net::{IpAddr, SocketAddr, TcpListener};
 use std::sync::Arc;
 use tower_http::compression::CompressionLayer;
+use tower_http::decompression::RequestDecompressionLayer;
 use tower_http::trace::TraceLayer;
 use utoipa::openapi::security::{Http, HttpAuthScheme, SecurityScheme};
 use utoipa::OpenApi;
@@ -144,6 +145,7 @@ impl HttpListenerParams {
             .deflate(true)
             .gzip(true)
             .quality(tower_http::CompressionLevel::Fastest);
+        let decompression_layer = RequestDecompressionLayer::new().deflate(true).gzip(true);

         let app = router_and_docs
             .router
@@ -167,6 +169,7 @@ impl HttpListenerParams {
                 auth_middleware,
             ))
             .layer(compression_layer)
+            .layer(decompression_layer)
             .layer(TraceLayer::new_for_http());
         let socket = TcpListener::bind(&self.listen)
             .with_context(|| format!("listen on {}", self.listen))?;

It also only explicitly enables only gzip/deflate to reduce the overall set of dependencies

[dschaaff]

It also only explicitly enables only gzip/deflate to reduce the overall set of dependencies

Thanks, I'll update this branch with that change.

...it looks similar but the layer is added more centrally:

Would you prefer I move the dependency to the kumo-server-common crate?